Multivariate DPOAE metrics for identifying changes in hearing: Perspectives from ototoxicity monitoring
Multivariate DPOAE metrics for identifying changes in hearing: Perspectives from ototoxicity monitoring
复制标题
DOI：
10.3109/14992027.2011.635713
复制
发表时间：
2012-02-01
影响因子：
2.7
通讯作者：
Dille, Marilyn F.
中科院分区：
文献类型：
--
作者：
Konrad-Martin, Dawn;Reavis, Kelly M.;Dille, Marilyn F.
畸变产物耳声发射（DPOAE）提供了一个窗口，实时耳蜗机械功能。然而，DPOAE指标的变化和听觉灵敏度之间的关系仍然知之甚少。阐明这些关系可能支持在听力保护计划（HCP）中使用DPOAEs检测导致噪声性听力损失（NIHL）的早期损伤，以便采取缓解措施限制任何持久的损伤。本报告描述了基于DPOAE的统计模型的发展，以评估癌症患者顺铂治疗的听力损失风险。使用机器学习范例构建耳毒性风险评估（ORA）模型，其中应用偏最小二乘法和留一法交叉验证，从一组已知的耳毒性风险因素和暴露前基线测量的DPOAE变化中产生最佳筛选算法。单DPOAE指标单独的耳毒性听力变化的风险比表现最好的多变量模型较差的指标。这一发现表明，在HCP中应用DPOAE的多变量方法将提高DPOAE测量在每个监测间隔识别噪声诱导的机械损伤和/或听力损失的耳朵的能力。这一预测必须在噪声暴露受试者中进行经验评估。
Distortion-product otoacoustic emissions (DPOAEs) provide a window into real-time cochlear mechanical function. Yet, relationships between the changes in DPOAE metrics and auditory sensitivity are still poorly understood. Explicating these relationships might support the use of DPOAEs in hearing conservation programs (HCPs) for detecting early damage leading to noise-induced hearing loss (NIHL) so that mitigating steps might be taken to limit any lasting damage. This report describes the development of DPOAE-based statistical models to assess the risk of hearing loss from cisplatin treatment among cancer patients. Ototoxicity risk assessment (ORA) models were constructed using a machine learning paradigm in which partial least squares and leave-one-out cross-validation were applied, yielding optimal screening algorithms from a set of known risk factors for ototoxicity and DPOAE changes from pre-exposure baseline measures. Single DPOAE metrics alone were poorer indicators of the risk of ototoxic hearing shifts than the best performing multivariate models. This finding suggests that multivariate approaches applied to the use of DPOAEs in a HCP, will improve the ability of DPOAE measures to identify ears with noise-induced mechanical damage and/or hearing loss at each monitoring interval. This prediction must be empirically assessed in noise-exposed subjects.